The Agent Skills Directory

[SAFE]: The skill serves as a high-level educational resource for designing agentic workflows. It contains architectural guidance, decision frameworks, and anti-patterns for managing Claude Code components.
[COMMAND_EXECUTION]: The documentation describes the use of platform features such as slash commands and hooks. It provides example Python and TypeScript code within markdown blocks to illustrate how users can implement security controls (e.g., blocking rm -rf commands) and observability dashboards. These are provided as educational templates rather than executable malicious scripts.
[DATA_EXFILTRATION]: Example scripts within the documentation demonstrate sending event summaries to a local observability server (http://localhost:3000). This is presented as a best practice for monitoring agent activity and does not constitute unauthorized data exfiltration.
[DYNAMIC_CONTEXT_INJECTION]: The skill documents the platform's syntax for executing shell commands within specific contexts (e.g., using the ! prefix). The documentation is descriptive and does not contain active shell injections intended to execute code during skill loading.

multi-agent-composition